The French government has published its first AI Usage Guide for state civil servants, an interministerial document aimed at nearly 2.5 million civil servants and designed to establish common rules on the use of these technologies in public administration. The document's goal is to bring order to a phenomenon that has already taken hold in its administrations. Civil servants use generative artificial intelligence in their daily work, but often without a clear framework determining what they can do, what they should avoid and what risks its use entails.
France regulates civil servants' everyday use of AI
The initiative has been driven by the Interministerial Directorate for Digital Affairs (DINUM), in collaboration with the Interministerial Directorate for Public Transformation (DITP) and the General Directorate for Administration and the Civil Service (DGAFP). The publication comes in a context in which generative AI is already part of the daily routine for a significant share of French public employees, with or without formal authorization.
According to Ishan Bhojwani, head of DINUM's AI department, many officials working in administrations that do not yet have their own charter or guide were calling for a clear framework that would allow them to distinguish which uses are permitted and which are not.
The concern is not merely theoretical. An internal study conducted during the first half of the year found that more than half of public officials were already using generative AI tools on their own initiative, often without supervision and without fully understanding the associated risks. This is the phenomenon known as shadow AI. That is, employees who turn to artificial intelligence tools outside the systems and procedures established by their organizations.
The guide itself makes its scope clear and presents itself as an interministerial reference document intended to standardize practices and support public officials, but without granting it normative or regulatory value in its own right. Rather than developing an extensive theoretical or legal framework, the document opts for an eminently practical approach, based on operational recommendations, concrete examples and points of attention that can be applied directly to civil servants' daily work.
Updating the guide periodically
The guide is expected to be updated periodically to adapt both to the evolution of artificial intelligence technologies and to changes in the legal framework, especially those deriving from the European Regulation on artificial intelligence and the General Data Protection Regulation. In the document's introduction, the Minister for Public Action and Accounts stresses that artificial intelligence no longer belongs to the future and is present in administrations, in the various public services and in the daily life of state bodies.
Javier Callejo warns about AI: the algorithm's verbal elegance weakens critical judgment
David Amiel also insists that AI will never replace the judgment of a well-trained public official aware of the scope of their decisions. Precisely because it does not replace it, he argues, artificial intelligence can become real support for civil servants' work, provided it is used correctly.
The guide places human responsibility at the center of any use of artificial intelligence within the public sector. No content generated by these systems can be used without prior verification adapted to its level of sensitivity. Caution must be especially high when AI is involved in an administrative decision, institutional communication, legal or financial analysis or the processing of personal data.
Equal treatment
The warning responds to known limitations of current artificial intelligence models: the possibility of generating errors or inaccurate information, the presence of biases, hallucination phenomena and a limited understanding of the real context of each situation.
The practical conclusion is that AI can assist, but the civil servant remains responsible for checking and validating the result. The French government also sets clear limits on how far the automation of administration can go. One of the main ones is that artificial intelligence will not make administrative decisions autonomously.
Nor may it replace professionals who maintain direct contact with citizens, such as teachers, healthcare staff, social workers or public-facing staff. According to official communications, this is both a regulatory and a political line in a particularly sensitive area, especially given the possibility of using automation to replace staff for budgetary reasons. For the government, that possibility is unacceptable.
The guide also reinforces another essential principle of public administration: equal treatment among citizens. The use of artificial intelligence cannot cause discrimination or break the principle of equality, an issue that is especially relevant after the criticism received by profiling algorithms used in the past to detect possible benefit fraud, a matter that remains pending judicial resolution.
The "Notre IA" plan
The guide is part of a broader plan called Notre IA, with which France aims to provide its administration with secure and sovereign artificial intelligence tools. Among them is L'Assistant, a conversational assistant developed by DINUM together with Mistral AI and hosted on infrastructure from Outscale, the cloud computing subsidiary of Dassault Systèmes.
The tool is already available to all state officials and allows tasks such as summarizing reports, rephrasing notes, extracting information from documents or translating texts. It has also recently added features that allow users to create projects based on their own document bases. The plan also includes tools aimed at the specific needs of certain bodies.
One of them is Langage Clair, used by staff at the General Directorate of Public Finances to rephrase administrative letters in simpler, more accessible language without losing precision. Specific applications are also planned for the Ministry of Justice for summarizing case files and transcribing hearings, with the aim of reducing deadlines and easing staff workload.
The French government starts from the premise that establishing rules is of little use if those who must apply them do not understand how the technology works. That is why the publication of the guide is accompanied by a training strategy. Starting in September, the future Public Service Institutes will incorporate artificial intelligence-related content into their initial and continuing training programs. The initiative adds to programs already available through the interministerial MENTOR platform.
These include a six-hour introductory course devoted to the concepts, challenges and opportunities of artificial intelligence, as well as a shorter program, developed together with the INRIA research institute, focused specifically on how generative AI works and on good practices for using it.
The strategy also draws on prior work by the DGAFP on human resources management, aimed at promoting ethical, people-centered use of artificial intelligence. Added to this are the recommendations issued by the Higher Commission for Digital Affairs and Postal Services on the use of artificial intelligence within the public sector.
Legitimate concerns about the future of public employment
The philosophy running through the document can be summed up in an idea defended by David Amiel himself: technology should serve to reduce the administrative burden on public employees, not to replace them. The strategy thus seeks to resolve a contradiction that public administrations are beginning to face directly.
On the one hand, generative AI offers tools capable of speeding up tasks, summarizing documents, rephrasing texts, processing information and reducing certain bureaucratic burdens. On the other, its use without controls introduces risks related to errors, biases, data protection, administrative liability and equal treatment. The government itself also acknowledges that the arrival of these technologies raises legitimate concerns about the future of public employment.
The guide and the training and support measures surrounding it are intended, at least in part, to address those concerns. With this first common framework, France's bet is not to expel artificial intelligence from the administration, but to determine where it can be used, under what conditions and, above all, where it must stop.
In fact, the principle running through the entire document is that AI can help the civil servant, but the decision and the responsibility must remain human.